Manufacturing Process Optimisation: A Conclusive Combined Study of Taguchi and Optimisation Technique With Artificial Intelligence

International Journal of Engineering Technology Science and Research
IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Manufacturing Process Optimisation : A Conclusive

Combined Study of Taguchi and Optimisation Technique
with Artificial Intelligence
B.RAMREDDY.
Associate Professor, Dept.Of Mech.Engg.G.N.D.Engg.College,Bidar.India
DR.D.CHOUDHARY
Professor,Dept.Of Mech.Engg.G.N.D.Engg.College,Bidar.India
ABSTRACT
Demand of products is increasing drastically which lead to growth of manufacturing industries. Recently, several
industries are getting established for satisfying the demand of products. However, there is a tradeoff between
manufacturing and quality of the product. Industrial growth depends on the quality of the product hence any industry
urge for development of good quality product. Manufacturing industries use conventional models for maintaining the
quality of product such as statistical measurement and employee involvement etc. but when multiple input configurations
are provided then it becomes a challenging task manual effort to provide better quality product. To overcome this,
Taguchi technique is presented for total quality management in manufacturing industries. In this work, we have
presented a conclusive study using combination of optimization technique and TQM technique which can improve the
performance of manufacturing process. First of all, TQM based model is applied for cartoon manufacturing industry
where punching machine, gluer, lamination and printing machine inputs are considered. These parameters fails to
optimize the machine downtime performance. To deal with this, Taguchi technique is applied which suggests the
parameter for improving machine downtime performance. In the next stage, TQM is tested for welding process
parameter optimization which is further improved by formulating L9 orthogonal array and neural network technique for
welding process parameter prediction. Finally, a combined approach of Taguchi, Particle swarm optimization and
neural network is presented which shows a significant improvement in the manufacturing process. Several experiments
are carried out and concluded that Taguchi technique helps to improve the manufacturing, moreover it can be optimized
and used along with artificial intelligence for improving the manufacturing efficiency.
KEYWORDS: Total quality management, Quality improvement, Taguchi technique, Performance, Arthogonal array,
Particle swarm optimisation, Artificial neural networks, Artificial intelligence.
1. INTRODUCTION
Industrial manufacturing process is growing rapidly which has led to the global competition in industries. In
order to meet the satisfactory manufacturing criteria, quality of the product plays an important role for
improvement of quality of product [1]. TQM (Total Quality Management) techniques have been adopted
widely because of its superior performance for improving the manufacturing quality [2]. Quality engineering
experts also have recommended to the other industries for implementing the TQM for time and money saving
purpose in manufacturing industries. Recently, industries have made efforts to make sure the employees
participation in quality improvement by reviewing conventional statistical and process control practices for
quality improvement. Conventional methods are not sufficient to obtain the desired manufacturing quality
management hence TQM is considered as most promising technique which can improve the performance of
the manufacturing systems. The main pillars of TQM are categorized into two main categories as continuous
quality improvement and obtaining good quality product by backing of all the concern. Several techniques
have been discussed recently which are focused on the quality improvement by including employees effort for
manufacturing [3].
1250 B.RAMREDDY, DR.D.CHOUDHARY

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Figure 1. Product manufacturing and lifecycle

According to the manufacturing process, product is processed through various stages for manufacturing which
is known as product development life cycle. This process is depicted in figure 1. This cycle contains various
stages where first of all, product specification are considered and analyzed. In next phase, CAD (computer
aided design) are used for design feature analysis, detailed design and prototype building. Furthermore,
Computer-Aided Manufacturing (CAM) is an important tool which can be used for quality assessment in
manufacturing process. By following all these processes such as drawing, prototype building, testing etc.,
However, for these processes, manual human effort is required which has more probability of error resulting
in degraded manufacturing quality.
NEED OF QUALITY MANAGEMENT IN MANUFACTURING INDUSTRIES

Conventional statistical techniques fail to provide desired performance for manufacturing monitoring and
error reduction. In order to address this, an improved technique is required which can help to improve the
manufacturing process. For manufacturing process, DrGenuchi Taguchi developed an online quality control
technique for industrial applications. This technique is considered as most promising technique due to its
significant nature for improving the manufacturing quality. Though a large number of case studies have been
reported [4] of the implementation of Taguchi’s offline quality control methods describing the conduct of a
minimum number of experiments to decide the optimum parameters, there is virtually no successful case
study in the literature about the implementation of Taguchi’s online quality control methods (hereafter shortly
referred to as TOLQC methods). However, the theoretical parts of TOLQC methods applied to hypothetical
case studies are described in [5]. From these studies, it can be concluded that still there is a need to develop an
efficient technique which can fill up the gap between Taguchi techniques and manufacturing product quality.
OVERVIEW OF TAGUCHI METHODS

Taguchi technique has been adopted widely for various manufacturing industries application. Several
researches are going on which are focused on the optimization of Taguchi technique. There are some
important features of taguchi technique, as discussed below:
 This technique is implemented at designing stage itself so that optimized design will be received for
further processing.
 According to this technique, rather than concentrating on inspection, quality concentration can improve
overall manufacturing.

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
 This process focuses on robust design concept due to which products don’t get disrupted during
production.
 Using statistical procedures in the design phase only, at appropriate stages, and completely removing them
from the production phase.
 Relying on the models developed based on the philosophy “loss to thesociety” rather than statistical
techniques during production phase.
Generally, this process is categorized into two main categories: (a) offline quality control and (b) online
quality control..According to offline process, Taguchi activities are employed during product design, product
planning and development stages. Taguchi introduced “design of experiments” method for offline quality
control techniques. According to offline quality control, orthogonal arrays are utilized which decides total
number of experiment, optimal parameter selection before proceeding towards actual production [6].
Other type of quality control technique is known as online technique for quality control. This technique is
different from conventional techniques of quality control. Conventional model for quality management
assumes that if product, process and services attain their full value as long as they fall between certain
deviations from the target. Online quality control is process where never ending cycle of quality improved is
applied to obtain the desired target [7]. Increasing demand of best quality product require an improvement in
Taguchi for best quality product manufacturing. Several techniques have been discussed where optimization
schemes are implemented and concluded that combination of optimization and Taguchi technique can
improve the manufacturing. In order to analyze the performance of Taguchi technique we have conducted 4
experimental studies as follows: (a) implementation of Taguchi technique for improvement of the production
process of cartoon manufacturing industry by considering four machines such as Printing, Punching, Gluer
and Lamination machines. (b) Welding Process Parameter optimization using Taguchi Method (c) prediction
of CO2 welding parameter using combined Taguchi L9 and neural network model. (d) PSO and Taguchi
technique for improving production quality using Neural Network.
Here we apply Taguchi technique for product quality management and tested for cartoon manufacturing
industry, welding process, CO2 welding process parameters and finally combination of particle swarm
optimization and Taguchi technique.
Rest of the article is organized as follows: section II deals with Taguchi technique implementation for cartoon
manufacturing industry. Similar technique is applied for welding process parameter optimization as discussed
in section III. Section IV describes a combined approach of neural network and Taguchi L9 for welding
parameter prediction and finally, a combined scheme of PSO and Taguchi method is presented for optimizing
Taguchi technique.
1. MANUFACTURING PROCESS ENHANCEMENT WITH THE HELP OF TAGUCHI

TECHNIQUE
As discussed before, in this section we present implementation of Taguchi technique for manufacturing
process enhancement. In this section, we discuss two important concepts of TQM and implementation of these
concepts for manufacturing process optimization. These concepts are as follows:
i) Continuous improvement of production process by applying newer and innovative methods
ii) Establishing performance measures for the processes.
In order to carry out this work, we have considered real-time experimental data from cartons manufacturing
company where we have considered four machines for experimental analysis such as Printing, Punching,
Gluer and Lamination machines. Initially, complete input and output details are collected and then specific
suggestions are given based on the all four machine inputs. At this stage, output values are measured in terms
of Availability, Performance, Quality and Overall Equipment Effectiveness which are found very less when
compared with world class industry. In order to improve the performance, machine down-time reduction is
suggested which can be obtained by TQM technique.

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
In this work, Overall equipment effectiveness (OEE) is considered as an important parameter which is used
for performance evaluation. OEE us a measurement of various components such as Availability, Performance,
and Quality. Moreover, performance of OEE is affected due to the machine downtime. Downtime is a time
duration where the machine system fails to execute elementary activities which leads to the performance
degradation. For further process, we have considered [8] for data collection of carton making industries. This
data contains multiple parameters of four machines. In order to reduce the downtime, total operating time
need to increased i.e. Availability=Total operating time/Total planned production time, (Performance=Total
pieces produced/operating time/production rate), quality (Quality=Total good pieces produced/Total pieces
produced) and OEE (OEE= Availability* Performance* Quality).The input constraints are total time(T),
Machine downtime(D), Total operating time (Tt), Total planned production time (Pt), Production breaks(B),
Production rate(R), Total no. of products produced(N), Total no. of products rejected(F), Total no. of high
quality products produced(G). Production breaks is set to default. Estimate certain input constraints according
to the equation (1)
= −
(1)
= −
2. WELDING PROCESS PARAMETER OPTIMIZATION
Previous section describes the implementation of Taguchi technique for improving the manufacturing process
for carton making industry. For more performance evaluation, we have considered another manufacturing
industry where welding process parameter is considered for obtaining better solution. Previous section
describes about advantages of Taguchi technique in manufacturing industries. Similar technique is discussed
and implemented for welding process parameter optimization by considering orthogonal array parameters. A
brief methodology is discussed here.
For any welding machine, Voltage, Electrode diameter, wire feed & nozzle to plate distance are known as the
most important parameters which are responsible for product quality. Furthermore, multiple combinations are
also analyzed to find the comparative analysis. Figure 1 shows CO2 welding process elements where
specimen is prepared using v-butt and a minimum gap between plates is maintained at 3 mm. Welding process
is performed by applying single run. CO2 gas is also used which works as shielded gas and rate of gas flow is
considered as 0.12 lit/min. with positive electrode polarity. Speed of welding torch was kept as 500 mm/min.
electrode wires are in the form of stainless steel along with copper coating. The chemical composition of test
specimen used was offollowing specifications: C-0.28%, Mn-1.25%, P-0.04%, S-0.05%, Yield strength=292
n/mm2 (min), Ultimate tensile strength=419N/mm2 (min).
Fig 2. Welding machine process

CO2 CVR-400 welding machine is used in this analysis. in complete setup, DC power source is also used with
constant voltage rectifier. Maximum torch speech is considered as 500 mm/min and forced air is also used for
cooling purpose [9]. The power source of three phase and 50 Hz is used

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
3. CO2 WELDING OUTPUT PARAMETERS USING TAGUCHI L9 ORTHOGONAL ARRAY

AND NEURAL NETWORK MODELS
Welding process parameters further can be improved by applying some artificial intelligence based scheme.
To do this, we have considered neural network technique for welding process parameters optimization. Use of
Taguchi technique has gained lot-of attraction from the researchers due to its significant nature to improve the
manufacturing process in industrial applications. Recent studies show that Taguchi can improve the
manufacturing, however, various researches have been presented recently which argue that the use of machine
learning or artificial intelligence can add more advantages to the system. Hence, for further optimization of
CO2 welding process, we apply neural network based learning approach which helps to predict the CO2
welding parameters.
As discussed in previous section, here also we have considered four main parameters which are used for
welding process optimization such as Voltage, Wire feed rate, Electrode diameter and Nozzle to work distance
to optimize three desired output variables. Output variables are known as penetration, width and
reinforcement. Hence, in this study, four input parameters and three output parameters are considered for
analysis. Two methods are chosen for optimization, namely, Orthogonal Array of Taguchi Method and the
Neural Network model. The Orthogonal method is a traditional method of optimization in the manufacturing
industry.
In manufacturing industries, multiple iterations are required for Taguchi methods to analyze the
manufacturing performance. However, according to Taguchi techniques, implementation cost and time both
are important which can be addressed by applying artificial intelligence scheme with Taguchi implementation.
This issue can be solved by reducing the number of experiments however it is always recommended to obtain
the desired performance in less computation time. In fact, Taguchi’s method exactly follows the same
principle and Taguchi defined the Orthogonal arrays to reduce the number of experiments as minimum as
possible. If a full factorial design is implemented, then huge number of experiments is required to derive the
transfer function for optimization. In this work, focus is made in the prediction of the output variable for any
arbitrary set of input variables rather than the optimization.
In this work, 4 factors are considered for prediction process where each of four variables can be set to the four
variables to three level or three process. For analysis, voltage values can be initialized from 20V, 27V or 34 V
similarly, other factors also can be maintained. The complete input, levels and codes are presented in the table
1. Each factor is coded as A, B, C and D. according to the conventional factorial design, total 81 simulation
runs required to obtain the desired performance.
Table1. Input levels and factors
Levels
SI. No. Code Factors
1 2 3
1 A Voltage[Volts] 20.0 27.0 34.0
2 B Wire feed rate[ mpm] 4.0 7.0 10.0
Electrode
3 C 0.8 1.0 1.20
Diameter[mm]
Nozzle to work
4 D 20.0 25.0 30.0
distance [mm]
It is common concept that running all 81 combinations becomes a complex and time consuming task, hence
Taguchi proposed L9 orthogonal array technique which can reduce the number of total iterations to obtain the
desired performance. The only combinations that are required to be tested is given in Table 2. According to
Taguchi, these are the combination of experiments that are sufficient to capture the behavior of the process
when the process variables change.

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Table 2: L9 orthogonal array

SI. No A B C D
1 1 1 1 1
2 1 2 2 2
3 1 3 3 3
4 2 1 2 3
5 2 2 3 1
6 2 3 1 2
7 3 1 3 2
8 3 2 1 3
9 3 3 2 1
ARTIFICIAL NEURAL NETWORK

In this section we discuss about artificial neural network and its implementation with TQM manufacturing
process optimization process. Neural networks are considered as significant component for financial analysis,
face recognition, prediction of rain fall and precipitation, prediction of traffic etc. Artificial neural network is
constructed based on the human brain process. In human brain various neurons are present which follows
parallel processing system. Neural network performs an important task called as weight computation. These
weights vary based on the input data type which are used for training the neural network. When the ANN is
trained with the input data along with the output data, then it is known as supervised learning [10]. If there are
additional hidden layers between the input and output neurons, then it is treated as multilayer perceptron
(MLP) networks. One of the most popular algorithms used in training the ANN is back propagation algorithm.
_ =∑ (1)
For Eq.1 it can be concluded that, each of the neuron received input from all the neurons from the previous
layer. Let the neuron is chosen in hidden layer, the _ is the total input to the neuron. There are
neurons in the layer preceding the hidden layer. is the weight of the connection between the neuron in
the hidden layer to a ith neuron in the layer that is proceeding the hidden layer. Each neuron in the ANN
produces an output by processing the inpout from the preceding layers using a transfer function. In this model,
a tangent hyperbolic function is used as given in Eq. 2.
= _
=1−
= ( − )
The ANN produces an output and the total error with respect to the desired output is computed using the
Eq. 3. The error E is minimized over all the outputs and over the required time series to optimize the weights
of the ANN.
4. COMBINED APPROACH OF PSO AND TAGUCHI TECHNIQUE FOR IMPROVING
PRODUCTION QUALITY USING NEURAL NETWORK
Previous sections have discussed about the importance of Taguchi technique in manufacturing process.
However, neural network based schemes are also implemented here which helps to improve the performance of
manufacturing. Further, we proposed a combined scheme of particle swarm optimization and neural network
implementation. This work is concentrated on optimization of surface roughness parameter resulting in quality
product manufacturing. Finally, we implement neural network approach to predict the desired surface
roughness parameters [11].

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Table 3 shows a chemical composition parameters for considered metal where total 12 metals are used for
analysis and their percentage use for analysis is also given here.
Table 3. Chemical composition for considered metal
S. No. Metal or % usage
element
1 Iron 98.38
2 Silicon 0.198
3 Vanadium 0.004
4 Sulphur 0.025
5 Manganese 0.761
6 Phosphorous 0.023
7 Aluminum 0.045
8 Molybdenum 0.012
9 Copper 0.056
10 Tungsten 0.045
11 Titanium 0.006
12 Carbon 0.436
Here, Taguchi technique based orthogonal array is developed by using 3 parameters such as peed of wheel,
feeding rate and depth of cut. For manufacturing process, time taken for analysis is considered as a challenging
task. Our aim is to reduce the total number of experiment runs. Minimum number of experiments can be
computed as:
= [( − 1) × ] + 1 (1)
Here values of and are considered as 3 hence equation (3) can be written as
= [(3 − 1) × 1] + 1 = 7 ≈ (2)
Figure 3 shows a general model for Taguchi technique. This process formulates a general problem where
multiple input parameters are provided for experimental study and analysis. Next stage, follows orthogonal
array formulation. After applying the experimental design, result analysis is presented by computing SNR
(Signal to Noise Ratio, ANOVA and response plot). Based on these parameters, outcome of Taguchi
technique is validated, if validation is true then the process is ended else experimental design is implemented
again to obtain the desired parameter.
Figure 3 Taguchi Technique

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
A. PARTICLE SWARM OPTIMIZATION

This subsection deals with particle swarm optimization approach The main advantage of this technique is that it
can be used for continuous or discrete optimization problem and gives better results in terms of optimization
performance. According to this study, each particle travels in dimensional space and tries to obtain best
position in the given search space.
Let us consider that two inputs of any given particles are given as and respectively where is the total
number of iterations. At this stage, our main aim is to obtain the best solution for each particle at iteration
denoted by . For best performance, global optimal solution need to be computed which can help to reduce the
optimization error. During this process, velocity of each particle can be updated as:
= + ∗ ∗ − + ∗
(1)
∗ −
Where velocity of particle at iterations denoted by , is the dimension of search space and is the
updated velocity , and are correlation coefficients and , random numbers which are ranging from 0 to
1, is the global best position in given dimensional subspace.
B. COMBINATION OF TAGUCHI AND PARTICLE SWARM OPTIMIZATION TECHNIQUE

In this section, a combination of Taguchi and particle swarm optimization is presented. According to Taguchi
model, orthogonal array is required for reducing the number of iterations effectively for given process.
Generally, according to Taguchi technique, orthogonal array model is given as ( , , , ) where denotes
number of rows, denotes total number of column or the parameters, denotes the level of orthogonal array
and strength is given by .
According to this approach, below mentioned steps are followed:
Step 1: Initialization of Taguchi Configuration
Step 2: PSO configuration parameter initialization
Step 3: Define Pareto Optimal solutions
Step 4: particle fitness computation
Step 5: initialization of personal and global best solution
Step 6: Factors and level computation for each particle
Step 7: Update Particle positions for each particle
Step 8: initialize Taguchi Swarm Generation and orthogonal array selection.
Step 9: New particle generation
Step 10: compute fitness and S/N ratio
Step 11: optima solution selection
In this process, we initate computation from 0 iteration for each particel in the given population by considering
velocity and and position parameters of each particle. Next stage belongs to generation of optimal solutions ,
fitness function computaiton for the initial population.
Here our main aim is minimize the surface roughness for the given material. For that particular parameter
Pareto Optimal solution is generated as follows:
1. Compute the objective vlues of feed rate, depth of cutting and speed of cutting
2. Sort the objective values as feed rate, depth of cutting and speed of cutting
3. Find an optimal Pareto soultion. If the obtained solution lies in between and then soluton is
considered as a Pareto Solution.

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Furthermore, we compute fitness function for given each particle. This parameter is is condered as the
adaptibility parameter for each population. If fitness function for any given population is high then there are
high chances for population adaptibility and survival of the population. In this work, the problem is considered
as a multi-objective problem. In order to perform the optimization we convert the problem into single objective
and a utility function or weight function is computed for solving the problem. Therefore , a fitness function is
defined by computing the weigthed sum for each particle given as:
( )= ( )+ ( )+ ( )+⋯
+ ( ) (2)
Where total numebr of objectives are considered for optimization which are given as , , and .
Figure 4 Taguchi and PSO implementation with optimal solution

According to proposed study, it follows the similar steps as mentioned in figure 4 but for further improvement
some extnesion is added into the conventional optimization. this approach follows same process as mentioned
before but it required two stages where the optimal solutions are obtained and best solution is selected from the
solutions. Finally we apply Artifcal Neural Network implementation for prediction the surface roughness
parameters. This decription is presented in next section.
C. NEURAL NETWORK IMPLEMENTATION

Here we present a general approahc which is used for industial application based on Taguchi optmization:
1. Provide input data such as speed of cutting, depth of cut and feed rate etc.
2. Genrate random conditions for cutting which can be changed later during simulation.
3. Surface roughness computation
4. Data prepration and divide to apply training and testing. Uniting of cutting conditions and other
calculated values into a data matrix. Normalization of the data in the matrix follows. All data must have their
values between 0–1. There is Breakdown of the data matrix into the input and output vector. Distribution of the
input/output vector into the two sets for training and testing.
Step 5: Use of ANN. The purpose of the neural network is to predict the surface roughness
Step 6: Selection of the type and architecture of the ANN and searching for optimum training parameters.
Step 7: Procedure of training of the ANN by using the training set.

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Step 8: Testing of trained ANN. If testing is successful and the error of prediction is within the permissible
limits, the empirical model is finished and ready for use. In case the testing is not successful, the training
procedure must be repeated with another larger set of training data or the training parameters must be changed.
5. EXPERIMENTAL RESULTS AND DISCUSSION

This section presents a brief simulation, experimental analysis and discussion. First of all, we present the
results of Taguchi technique implementation for cartons manufacturing industry as our first objective.
According to first study, down time analysis study is presented where analysis is given for all considered four
machines. First of all, printing machine analysis are presented:
Down time Frequency of

reasons occurrence
Machine 18
cleaning
Color approval 14
All blanket 105
cleaning*
Ink tap roll 21
cleaning
Plate fix and 42
make ready
Coating blanket 24
fixing
Whole ink dock 12
cleaning
Roller washing 19
All unit blanket 20
fixing
Cylinder 9
cleaning
Roller change 7
Water change 3
and maintenance
Figure 5. Offset printing machine performance
The fig (5) describes the variation of blanket cleaning process. Actually the blanket cleaning process is the
one of the major time consuming and frequently needed process of downtime. As a innovative technique of
process optimization and in order to rectify the process the new suggestion of introducing the twin turbo
system for cleaning the printing machine (the*marked) downtime will be reduced automatically. Because it
( )
reduces the blanket cleaning process up to 90%. Total downtime percentage analysis computed is =( . )
=
67.6. Therefore the downtime in min.(i.e. Reduced downtime) =255*68% = 173min. which is tabulated in
table (1).

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Similarly, down time variation analysis is computed for each machine. Punching Machine total downtime in
%ge=(100-37) +(37-29.6)=63+7.4=70.4Therefore the total downtime in min.(i.e.
reduceddowntime)=150*70%=105min. which is tabulated in table 2. Gluer Machine the total downtime in
%ge=(100-22)+(22- 17.6)=78+4.4=82.4.
Hence the downtime in min.(i.e. reduced downtime)=30*83%=25 . Lamination Machine total downtime in
%ge=(100-9)+(9-7.2)=91+1.8=93(approx).
Therefore the total downtime in min.(i.e reduced downtime)=90*93%=84.
With the help of these techniques, we obtain the suggested parameters to obtain the best manufacturing.
Suggested results for printing machine are given in below mentioned table.
Table 4. Printing machine downtime analysis.
Production/shift 1 2 3 4 5 6 7 8 9 10
Total time(minutes) 510 510 720 510 720 510 690 690 690 750
Production Breaks 30 30 30 30 30 30 0 0 0 30
(minutes)
Machine 173 184 265 235 235 204 306 275 275 286
Downtime(minutes)
Production Rate 266 266 266 266 266 266 266 266 266 266
(Unit/minute)
Total Cartoons 36830 27243 37872 26524 45507 47353 30439 42942 31461 70998
produced
Total cartoons 1473 1090 1515 1061 1820 1894 1218 1718 1285 2840
rejected
Total good cartoons 35357 26153 36357 25463 43687 45459 29221 41224 30203 68158
produced
Total planned 480 480 690 480 690 480 690 690 690 720
production time
Total operating time 307 296 425 245 455 276 384 415 415 434
Similarly, punching machine down-time analysis and suggested parameters are given in table (5)
Table 5. Punching machine downtime analysis
(minutes)
Machine 105 210 273 210 242 147 158 74 53 63
Downtime(minutes)
Production Rate 70 70 70 70 70 70 70 70 70 70
(Unit/minute)
Total Cartoons 23184 16229 11592 5425 20325 25385 18219 25237 27857 21863
produced
Total cartoons 116 81 58 27 102 127 91 126 139 109
rejected
produced
Total planned 450 450 480 480 480 480 480 480 480 480
production time
Total operating time 345 240 207 270 238 333 322 406 427 417

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Gluer machine suggested parameters are given in table (6)

Table 6. Gluer machine down time analysis.
(minutes)
Machine 25 271 295 172 148 98 418 197 344 172
Downtime(minutes)
Production Rate 350 350 350 350 350 350 350 350 350 350
(Unit/minute)
Total Cartoons 50164 97582 54194 79125 67861 139981 192457 67354 61214 26964
produced
Total cartoons 552 1073 596 870 746 1540 2117 741 673 297
rejected
produced
Total planned 480 480 690 480 480 720 780 480 480 600
production time
Total operating 455 209 395 308 332 622 362 283 136 428
time
Lamination machine suggested results are:

Table 7. Lamination machine downtime analysis.
(minutes)
Machine 84 112 84 251 56 223 112 28 84 112
Downtime(minutes)
Production Rate 26 26 26 26 26 26 26 26 26 26
(Unit/minute)
Total Cartoons 7022 7951 8772 4692 8753 5780 9751 9195 12131 15742
produced
Total cartoons 77 87 96 52 96 64 107 101 133 173
rejected
produced
Total planned 480 480 690 480 480 480 630 450 660 960
production time
Total operating 396 368 396 229 424 257 518 422 576 848
time
Finally, combined performance for considered input parameters is given in the table below in table (8):

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Table 8. comparative analysis for each machine.

m/c Printing m/c Punching m/c Gluer m/c Lamination
(shift no.) (shift no.) (Shift no.) m/c(shift no.)
Availability Real 46.9211 (1) 84.366(9) 93.8116(1) 93.2054 (8)
(%) Suggested 63.95833 (1) 88.9583 (9) 94.79167 (1) 93.77(8)
Performance Real 64.5436(6) 108.85 (6) 133.373 (2) 86.524 (6)
(%) Suggested 64.9 (6) 109.6 (6) 135 (2) 88.1 (6)
Quality (%) Real 96.09 (6) 99.594 (3) 99.9931 (9) 98.992(1)
Suggested 97.02(6) 99.9 (3) 99.6 (9) 99.3 (1)
OEE (%) Real 24.5765 (10) 78.29 (9) 51.97 (7) 77.3727 (8)
Suggested 36.65 (10) 84.044 (9) 70.680 (7) 78.99 (8)
This analysis shows that Taguchi based technique can improve the performance of manufacturing industries.
In next study, we study about welding parameter optimization where input parameters are taken from
configuration given in table 1. For this analysis, orthogonal array is formulated which is presented in table 2.
Table 9. Experiment layout
Input parameters output on profile projector
Exp. No. A [volt.] B [mpm] C [mm] D [mm] Penetration Width Reinforcement
[mm] [mm] [mm]
1 20 4 0.8 20 3.75 9 2.2
2 20 7 1.0 25 4 9.85 2.90
3 20 10 1.2 30 5.5 10.55 1.5
4 27 4 1.0 30 5.2 11.9 2
5 27 7 1.2 20 5.3 10.5 1.7
6 27 10 0.8 25 5 11.15 2.1
7 34 4 1.2 25 5 11.15 1.70
8 34 7 0.8 30 5.40 9.25 1.80
9 34 10 1 20 5.90 12.20 2.40
Above given table shows experimental layout where input and output parameters are presented divided into
factors and experiment numbers. This study is validated by considering higher-is better concept based S/N
calculation. This study is presented in table 10 as given below.
Table .10 . S/N ratio calculation for penetration, reinforcement and width (higher is better)
Exp. No. Penetration Reinforcement Width
1 11.48 6.84 19.08
2 12.04 9.24 19.86
3 14.80 3.52 20.46
4 14.32 6.02 21.51
5 14.48 6.84 20.42
6 13.97 4.60 20.94
7 12.25 6.44 19.55
8 14.64 5.10 19.32
9 15.41 7.60 21.72

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Next test case includes neural network and Taguchi technique combination for welding process parameter
estimation. According to this study, L9 orthogonal array is considered for welding process parameters,
predicted parameters are obtained using neural network and Taguchi as depicted in below given table (11).
Table 11. Neural network and taguchi predicted parameters.
PREDICTION 1 2 3 4 5
NO.
AND LEVELS
A 2 2 2 1 3
B 3 2 3 3 1
PENETRATIO FACTOR
CODES
C 2 1 1 1 3
D 3 3 3 3 1
Taguchi
6.32 5.43 6.0 5.25 4.72
L9
Neural
N (mm)
Network 5.65 4.71 4.75 4.67 4.33
Taguchi
L9 13.07 10.12 11.55 10.17 9.90
REINFORCEMEN WIDTH
(mm)
Neural
11.23 10.60 10.59 10.09 9.15
Network
Taguchi 1.77
1.67 1.23 1.47 2.13
L9
Neural
Network
1.59 1.99 1.64 1.88 2.09
(mm)
T
First of all, chemical composition of metal is analyzed as depicted in table 11. For experimental study, we have
considered selected parameters which are given in table 12.
Table 12. Generalized Chemical composition
Parameter name C Fe Mn P S Others
Parameter value 0.37 98.8 0.7 0.035 0.045 0.05
Similarly, considered mechanical parameters are presented in table 13.

Table 13. Mechanical properties
Ultimate Tensile strength ( ) 515
Yield Tensile strength 450

Elasticity 200
Hardness Vickers 155

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Furthermore, cutting conditions which are used in this work are given in table 14.
Table 14. Cutting Condition parameters
Speed of cutting ( ) 560 640 960
Feed rate ( ) 0.16 0.17 0.20
Depth of cutting ( ) 0.2 0.3 0.4
In order to analysis the performance of the proposed approach, we have considered 9 specimens with the help
of cutting condition combination. This combination is presented in table 15.
Table 15. Cutting parameters for considered specimen
Specimen No. ( / ) ( )
1 560 560 640

2 560 0.16 0.17
3 560 0.2 0.3
4 640 0.16 0.3
5 640 0.17 0.4
6 640 0.2 0.2
7 960 0.16 0.4
8 960 0.17 0.2
9 960 0.2 0.3
In this work our main aim is to provide an efficient approach for design optimization using Taguchi, PSO and
artificial neural network based techniques. Here we have presented a hybrid model for performance
improvement. As discussed in previous section that the multi-objective optimization problem is transformed
into single objective therefore we have considered surface roughness parameter for optimization and it is used
in Taguchi, PSO and ANN. This parameter is obtained from 9 specimens as depicted in table 15. Total number
of specimen considered are 27 but here we have used 9 specimen and rest of 18 are predicted using neural
network prediction model.
From previous studies, we have concluded that Taguchi techniques are significant techniques which helps to
obtain high-quality manufacturing systems. It reduces manufacturing cost and improves the quality of products.
In order to perform this operation, orthogonal array is used which helps to analyze the parameters and assigns
minimum number of required experiments. Taguchi technique reduces error in experiments by applying
random ordering of experiments. This technique analyzes means response and signal to noise ratio in each
experiments. Further, ANOVA (Analysis of variance) is also used in this technique which analyses designing
factor and interaction which affects the output response.
However, in this work, main aim is to apply Taguchi technique for manufacturing quality assurance. Hence,
first of all Taguchi technique is applied for quality management service. Furthermore, an optimization strategy
is presented for performance improvement by considering quality and cost as the key constraints.
In this work, we consider 27 specimen considering depth of cut, cutting speed and metal feed rate. These
constraints are considered as attributes. For further improvement, we apply Particle Swarm Optimization
technique along with Taguchi technique with neural network prediction modeling. Neural network helps to
obtain the efficient prediction for various attributes. Here surface roughness is considered for optimization for

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
multiple levels of operations. According to Taguchi technique, suitable factors can be selected based on their
levels of experimental analysis where three factors are considered for all three levels are presented in table 16.
Table 16. Factor level representation
Cutting conditions Factor for Level 1 Factor for level 2 Factor for level 3
( ) 0.2 0.3 0.4
0.16 0.17 0.20
/min) 560 640 960

Later, an orthogonal array 9 is taken as presented in table 17.
Experiment ( ( ) A B C Output
( )
number / ) Response
of surface
roughness
1 560 0.16 0.2 1 1 1 3.4
2 560 0.17 0.3 1 2 2 3.35
3 560 0.20 0.4 1 3 3 4.42
4 640 0.16 0.3 2 1 2 3.3
5 640 0.17 0.4 2 2 3 3.95
6 640 0.20 0.2 2 3 1 3.90
7 960 0.16 0.4 3 1 3 3.27
8 960 0.17 0.2 3 2 1 2.87
9 960 0.20 0.3 3 3 2 2.48
Generally, Taguchi technique depends on the below mentioned steps:

 Identification of performance response for optimization
 Estimation of number of level for each parameter
 Orthogonal array selection
 Perform random experiment for given orthogonal array
 Compute mean response and signal to noise ratio
 Apply ANOVA analysis
 Optimal parameter selection
After applying this stage, Particle swarm optimization step is applied which is an iterative process. This
optimization is applied on the surface roughness and for each experiment, a separate optimal value is obtained.
These two techniques i.e. Taguchi and PSO are applied for 9 specimens, remaining experiment response is
predicted using artificial neural network. the artificial neural network uses back-propagation model with
sigmoid activation function. At this stage, 9 specimens are considered for training purpose and remaining
specimens are used for testing. The output of neural network prediction is considered as predicted surface
roughness.
ANOVA analysis for given cutting condition parameters is performed as presented in table 18. Here inner array
output is arranged in the last column of the table where surface roughness and signal-to-noise ratio is
considered as output and presented in outer array.

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Table 18. Signal to Noise ratio analysis
Cutting condition Factors S/N
No. of ( ) ( ) A B C
trial
1 560 0.161 0.2 1 1 1 3.4 -10.76
2 560 0.17 0.3 1 2 2 3.35 -10.51
3 560 0.2 .4 1 3 3 4.42 -12.91
4 640 0.16 0.3 2 1 2 3.3 -10.56
5 640 0.17 0.4 2 2 3 3.95 -11.98
6 640 0.20 0.2 2 3 1 3.90 -11.80
7 960 0.16 0.4 3 1 3 3.27 -10.29
8 960 0.17 0.2 3 2 1 2.87 -9.16
9 960 0.20 0.3 3 3 2 2.48 -7.92
Later we compute the average effect of the output response of surface roughness and signal-to-noise ratio for
each factor of the experiment in each level. The outcome is presented in the table 19. For each experiment is
computed as: A= 3.4 +3.35 + 4.42/3=3.72. For parameter B it can be computed as: roughness at train 1 +
roughness at experiment 4+ roughness at experiment 7= 3.4+3.3+3.27=3.36 and similarly, for depth of cut:
(3.4+ 3.90 + 2.87) / 3 = 3.39
Table 19. Effect of surface roughness

Cutting Level 1 Level 2 Level 3 Difference Rank
condition
3.72 3.74 2.85 0.89 1
3.36 3.40 3.61 0.25 3
3.39 3.069 3.85 0.781 2
Finally, a comparative analysis of surface roughness is presented in table 20. In this table we show output
response using Taguchi technique and proposed hybrid technique for 18 specimens.

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
Table 20. Comparative analysis

Sl. No. ( / ) ( ) Taguchi Proposed
( )
1 560 0.16 0.2 3.60 3.45
2 560 0.17 0.3 3.303 3.34

3 560 0.20 0.4 4.312 4.40
4 640 0.16 0.3 3.274 3.39
5 640 0.17 0.4 4.121 3.99

6 640 0.20 0.2 3.841 3.91
7 960 0.16 0.4 3.212 3.22

8 960 0.17 0.2 2.762 2.85
9 960 0.20 0.3 2.633 2.46
10 560 0.16 0.3 3.278 3.45

11 560 0.16 0.4 4.087 3.98
12 560 0.17 0.2 3.651 3.44
13 560 0.17 0.4 4.153 4.31
14 560 0.2 0.2 3.851 3.79
15 560 0.2 0.3 3.492 4.33
16 960 0.2 0.4 3.44 3.65
17 960 0.17 0.2 2.96 2.90
18 960 0.2 0.4 3.44 3.50
Table 20 shows that proposed hybrid approach provide significant performance with the help of Artificial
Neural Network prediction scheme.
6. CONCLUSION
In this work, a combined experimental study is presented for improving the quality of manufacturing. In order
to perform this task, well-known and widely adopted scheme of quality management, Taguchi TQM (Total
Quality Management) is considered. For experimental study, first of all we have taken data from cartoon
manufacturing industry where punching, lamination, gluer and lamination machines are considered along with
their input and output parameters. These output parameters are not meeting the standard criteria in terms of
machine downtime, for efficient manufacturing, hence to improve this Taguchi technique is applied which
helps to improve the manufacturing process by reducing machine downtime.
Further, we have considered CO2 welding for analysis, where Taguchi is applied for welding process
parameter optimization. However, this approach is further processed for CO2 welding parameter prediction

IJETSR
www.ijetsr.com
ISSN 2394 – 3386
Volume 5, Issue 3
March 2018
using combined Taguchi and neural network process. Experimental study shows that, combined scheme can
improve the performance of the model. Finally, we have proposed a hybrid technique for improving the
manufacturing quality where Taguchi technique is combined with PSO (Particle Swarm Optimization) and
neural network based prediction model is applied for surface roughness parameter prediction. Experimental
study shows that Taguchi technique provides desired input parameters for improving the performance but it
can be improved further by using optimization
7. REFERENCES
1. Inman, Robert R., et al. "Survey of recent advances on the interface between production system design and
quality." IIE transactions 45.6 (2013): 557-574.
2. Bayazit, Ozden, and BirsenKarpak. "An analytical network process-based framework for successful total quality
management (TQM): An assessment of Turkish manufacturing industry readiness." International Journal of
Production Economics 105.1 (2007): 79-96.
3. Dhillon, B.S., “Quality circle: bibliography”, International Journal of Reliability & Quality Management, Vol. 5 No.
1, 1988, pp. 53-75.
4. Bendell, T. (Ed.), Taguchi Methods – Proceedings of the 1988 European Conference, Elsevier Applied Science,
London, 1988.
5. Taguchi, G., Elsayed, E.A. and Hsiang, T., Quality Engineering in Production Systems, McGraw-Hill, New York,
NY, 1989
6. Kamal, Alaa, et al. "Power Consumption Optimization Based on B-MAC Protocol for Multi-Scenario WSN by
Taguchi Method." International Conference of Reliable Information and Communication Technology. Springer,
Cham, 2017.
7. W. C. Weng, F. Yang, and A. Z. Elsherbeni, Electromagnetics and Antenna Optimization using Taguchi's Method,
Morgan & Claypool Publishers, CA, December 2007
8. Pradeep Kumar, K. V. M. Varambally and Lewlyn L.R.Rodrigues, “A Methodology for Implementing Total
Productive Maintenance in Manufacturing Industries–A Case Study”, International Journal of Engineering Research
and Development,Vol.5,No.2,pp. 32-39,2013
9. "Effect of Welding Process Parameters on Weld Bead Geometry and Optimization of Process Parameters
toMaximize Depth to Width Ratio for Stainless Steel Gas Tungsten Arc Welded Plates Using Genetic
Algorithm",European Journal of Scientific Research, Vol. 62, No. 1 (2011), pp. 76-94
10. Benardos, P. G., and G-C. Vosniakos. "Optimizing feedforward artificial neural network architecture." Engineering
Applications of Artificial Intelligence20.3 (2007): 365-382.
11. Argha Das, ArindamMajumder, Pankaj Kr. Das, Detection of Apposite PSO Parameters Using Taguchi Based Grey
Relational Analysis: Optimization and Implementation Aspects on Manufacturing Related Problem, Procedia
Materials Science, Volume 6, 2014, Pages 597-604, ISSN 2211-8128

Manufacturing Process Optimisation: A Conclusive Combined Study of Taguchi and Optimisation Technique With Artificial Intelligence

Caricato da

Informazioni sul documento

Descrizione originale:

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Manufacturing Process Optimisation: A Conclusive Combined Study of Taguchi and Optimisation Technique With Artificial Intelligence

Caricato da

Copyright:

Formati disponibili

International Journal of Engineering Technology Science and Research

Manufacturing Process Optimisation : A Conclusive

1250 B.RAMREDDY, DR.D.CHOUDHARY

Figure 1. Product manufacturing and lifecycle

NEED OF QUALITY MANAGEMENT IN MANUFACTURING INDUSTRIES

OVERVIEW OF TAGUCHI METHODS

1251 B.RAMREDDY, DR.D.CHOUDHARY

1. MANUFACTURING PROCESS ENHANCEMENT WITH THE HELP OF TAGUCHI

1252 B.RAMREDDY, DR.D.CHOUDHARY

Fig 2. Welding machine process

1253 B.RAMREDDY, DR.D.CHOUDHARY

3. CO2 WELDING OUTPUT PARAMETERS USING TAGUCHI L9 ORTHOGONAL ARRAY

1254 B.RAMREDDY, DR.D.CHOUDHARY

Table 2: L9 orthogonal array

ARTIFICIAL NEURAL NETWORK

1255 B.RAMREDDY, DR.D.CHOUDHARY

Figure 3 Taguchi Technique

1256 B.RAMREDDY, DR.D.CHOUDHARY

A. PARTICLE SWARM OPTIMIZATION

B. COMBINATION OF TAGUCHI AND PARTICLE SWARM OPTIMIZATION TECHNIQUE

1257 B.RAMREDDY, DR.D.CHOUDHARY

Figure 4 Taguchi and PSO implementation with optimal solution

C. NEURAL NETWORK IMPLEMENTATION

1258 B.RAMREDDY, DR.D.CHOUDHARY

5. EXPERIMENTAL RESULTS AND DISCUSSION

Down time Frequency of

1259 B.RAMREDDY, DR.D.CHOUDHARY

1260 B.RAMREDDY, DR.D.CHOUDHARY

Gluer machine suggested parameters are given in table (6)

Lamination machine suggested results are:

1261 B.RAMREDDY, DR.D.CHOUDHARY

Table 8. comparative analysis for each machine.

1262 B.RAMREDDY, DR.D.CHOUDHARY

Network 5.65 4.71 4.75 4.67 4.33

Parameter value 0.37 98.8 0.7 0.035 0.045 0.05

Similarly, considered mechanical parameters are presented in table 13.

Yield Tensile strength 450

1263 B.RAMREDDY, DR.D.CHOUDHARY

Feed rate ( ) 0.16 0.17 0.20

Depth of cutting ( ) 0.2 0.3 0.4

1 560 560 640

1264 B.RAMREDDY, DR.D.CHOUDHARY

/min) 560 640 960

Generally, Taguchi technique depends on the below mentioned steps:

1265 B.RAMREDDY, DR.D.CHOUDHARY

Table 18. Signal to Noise ratio analysis

Cutting condition Factors S/N

1 560 0.161 0.2 1 1 1 3.4 -10.76

2 560 0.17 0.3 1 2 2 3.35 -10.51

3 560 0.2 .4 1 3 3 4.42 -12.91

4 640 0.16 0.3 2 1 2 3.3 -10.56

5 640 0.17 0.4 2 2 3 3.95 -11.98

6 640 0.20 0.2 2 3 1 3.90 -11.80

7 960 0.16 0.4 3 1 3 3.27 -10.29

8 960 0.17 0.2 3 2 1 2.87 -9.16

9 960 0.20 0.3 3 3 2 2.48 -7.92

Table 19. Effect of surface roughness

3.72 3.74 2.85 0.89 1

3.36 3.40 3.61 0.25 3

3.39 3.069 3.85 0.781 2

1266 B.RAMREDDY, DR.D.CHOUDHARY

Table 20. Comparative analysis